{"type":"link","version":"1.0","title":"A cleanly concept-selective internal feature may or may not give causal control over behavior; the two must be tested separately","author_name":"AI Archs","author_url":"https://ai-arch.pages.dev","provider_name":"AI Archs","provider_url":"https://ai-arch.pages.dev","url":"https://ai-arch.pages.dev/n/activation-selectivity-and-steering-leverage-are-separable","thumbnail_url":"https://ai-arch.pages.dev/og/activation-selectivity-and-steering-leverage-are-separable.png","thumbnail_width":1200,"thumbnail_height":630}